TREE-G: Decision Trees Contesting Graph Neural Networks
Authors: Maya Bechler-Speicher, Amir Globerson, Ran Gilad-Bachrach
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted experiments using TREE-G on graph and vertex labeling benchmarks and compared its performance to popular GNNs, graph kernels, and other tree-based methods. [...] The results are summarized in Tables 1, 2 & 3. |
| Researcher Affiliation | Academia | Maya Bechler-Speicher1, Amir Globerson1, Ran Gilad-Bachrach2 1 Blavatnik School of Computer Science, Tel-Aviv University 2 Department of Bio-Medical Engineering and Edmond J. Safra Center for Bioinformatics,Tel-Aviv University |
| Pseudocode | Yes | Algorithm 1: Infer TREE-G Data: A: adjacency matrix, X: node features matrix, i: node index for inference in the case of vertex labeling task, N: a TREE-G node with stored parameters kN, dN, N, ρN, rN, θNand AGGNin the case of graph labeling tasks Result: label y Function Infer Vertex(N, A, X, i): [...] Algorithm 2: Train TREE-G Data: D: training dataset of labeled graphs or labeled nodes. Result: Root of a trained TREE-G T Pre-Defined: Optimization criterion (e.g., Gini score or the L2 loss), stop conditions (e.g., minimal gain, maximal depth, minimal examples in leaf), label computation rule (e.g., majority vote or average of labels) Function Train TREE-G(D): |
| Open Source Code | Yes | 6Code is available at github.com/mayabechlerspeicher/TREE-G |
| Open Datasets | Yes | Graph Prediction Tasks: We used nine graph classification benchmarks from TUDatasets (Morris et al. 2020), and one large-scale dataset from the Open Graph Benchmark (Hu et al. 2020). [...] Node Prediction Tasks: We used the three citation graphs tasks Cora, Citeseer & Pubmed from Planetoid (Yang, Cohen, and Salakhutdinov 2016). |
| Dataset Splits | Yes | For graph prediction tasks we report the average accuracy and std of a 10-fold nested cross-validationn, except for the mol HIV dataset for which we use official pre-defined splits provided in Hu et al. (2020) and the metric is AUC. For the node prediction tasks we report average accuracy and std using the pre-defined splits provided in the data, except for the county dataset, which is a regression task, and hence we use RMSE instead of accuracy. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) are provided for the experimental setup. |
| Software Dependencies | No | The paper does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions or specific library versions). |
| Experiment Setup | No | More details on the evaluation protocol including the tuned hyper-parameters for each algorithm are provided in the Appendix. |